Modeling framework and comparison of memristive devices and associated STDP learning windows for neuromorphic applications

Abstract

This paper presents a comparative synthesis of the suitability of three memristive device technologies and their corresponding spike-timing-dependent plasticity (STDP) learning windows for neuromorphic applications. The physical mechanisms behind the nonlinear switching memristive dynamics of ReRAM, based on titanium dioxide, ferroelectric tunnel junctions, and phase change memory are analyzed towards the development of accurate and computationally efficient compact models which are implemented as a Verilog-A description. The developed Verilog-A compact models are separately validated and compared with the measurement data. Moreover, the asynchronous STDP learning rule is implemented using the above mentioned memristive devices as artificial synapse for spike-based neuromorphic computing. The considered memristive technologies are compared and discussed towards their integration in fast and/or large-scale circuit implementations.